TTS Arena - Bwrdd Canlyniadau Modelau Lleferydd AIName
Cymharu mwy na 20 model testun-i-leferydd. Benchmarks swyddogol, graddau'r gymuned, a chymharu ochr yn ochr.
Cymhariaeth ochr- wrth- ochr
Teipiwch destun, dewiswch ddau ddull, a chymharu'r canlyniadau. Nid oes angen cyfrif ar ddull am ddim.
Mae modelau am ddim yn gweithio heb gyfrif. Cofrestru to compare premium models.
Model Bwrdd Arwain
| # | Model | Swyddogol | Cymuned | Eich gradd | Cyflymder | o Fawrth |
|---|---|---|---|---|---|---|
| 1 |
Kokoro
Lightweight 82M parameter model delivering studio-quality speech with blazing-fast inference.
82M
1200h
2024
|
4.8 /5 |
5.0
/5
1 pleidlais
|
fast | Free | |
| 2 |
CosyVoice 2
Alibaba's scalable streaming TTS with human-parity naturalness and near-zero latency.
300M
200000h
2024
|
4.26 /5 | Dim pleidleisiau eto | medium | Standard | |
| 3 |
Chatterbox
State-of-the-art zero-shot voice cloning with emotion control from Resemble AI.
300M
2025
|
4.25 /5 | Dim pleidleisiau eto | medium | Premium | |
| 4 |
StyleTTS 2
Human-level text-to-speech through style diffusion and adversarial training.
100M
585h
2024
|
4.23 /5 | Dim pleidleisiau eto | medium | Premium | |
| 5 |
Piper
A fast, local neural text to speech system optimized for Raspberry Pi and embedded devices.
15M
2023
|
4.15 /5 | Dim pleidleisiau eto | fast | Free | |
| 6 |
MeloTTS
High-quality multilingual text-to-speech that runs on CPU with minimal latency.
25M
2024
|
4.13 /5 | Dim pleidleisiau eto | fast | Free | |
| 7 |
Dia TTS
Multi-speaker dialog generation model that creates natural conversations between speakers.
1.6B
2024
|
4.09 /5 | Dim pleidleisiau eto | medium | Standard | |
| 8 |
VITS
Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech.
25M
585h
2021
|
4.0 /5 | Dim pleidleisiau eto | fast | Free | |
| 9 |
Orpheus
Human-level emotional TTS model trained on 100K hours of speech data.
3B
100000h
2025
|
4.0 /5 | Dim pleidleisiau eto | medium | Standard | |
| 10 |
OpenVoice
Instant voice cloning with granular control over style, emotion, and accent.
300M
2024
|
4.0 /5 | Dim pleidleisiau eto | medium | Premium | |
| 11 |
IndexTTS-2
Zero-shot TTS with fine-grained emotion control and high expressiveness.
300M
2025
|
3.91 /5 | Dim pleidleisiau eto | medium | Standard | |
| 12 |
Spark TTS
Voice cloning TTS with controllable emotion and speaking style via prompts.
500M
2025
|
3.9 /5 | Dim pleidleisiau eto | medium | Standard | |
| 13 |
Parler TTS
Describe the voice you want in natural language and Parler generates matching speech.
880M
45000h
2024
|
3.83 /5 | Dim pleidleisiau eto | medium | Standard | |
| 14 |
Tortoise TTS
Multi-voice text-to-speech focused on quality with autoregressive architecture.
400M
50000h
2022
|
3.7 /5 | Dim pleidleisiau eto | slow | Premium | |
| 15 |
Bark
Transformer-based text-to-audio model that generates realistic speech, music, and sound effects.
350M
100000h
2023
|
3.57 /5 | Dim pleidleisiau eto | slow | Standard | |
| 16 |
Bark Small
Lighter version of Bark with faster inference and lower memory usage.
150M
100000h
2023
|
— | Dim pleidleisiau eto | medium | Standard | |
| 17 |
GPT-SoVITS
Few-shot voice cloning TTS that replicates any voice from just 5 seconds of audio.
200M
2024
|
— | Dim pleidleisiau eto | slow | Standard | |
| 18 |
Qwen3 TTS
Alibaba's multilingual TTS with voice cloning, preset voices, and voice design from text.
1.7B
2025
|
— | Dim pleidleisiau eto | medium | Standard |
Sgorau Benchmark Manwl
Sgoriau cyfartalog swyddogol TTS.ai ar draws tri dimensiwn: naturioldeb, cywirdeb, a chyflymder.
Kokoro
Free
Naturioldeb
4.8/5
Cywirdeb
4.7/5
Cyflymder
4.9/5
Cyfanswm
4.8/5
CosyVoice 2
Standard
Naturioldeb
4.5/5
Cywirdeb
4.4/5
Cyflymder
3.8/5
Cyfanswm
4.26/5
Chatterbox
Premium
Naturioldeb
4.7/5
Cywirdeb
4.5/5
Cyflymder
3.4/5
Cyfanswm
4.25/5
StyleTTS 2
Premium
Naturioldeb
4.5/5
Cywirdeb
4.3/5
Cyflymder
3.8/5
Cyfanswm
4.23/5
Piper
Free
Naturioldeb
3.5/5
Cywirdeb
4.2/5
Cyflymder
4.95/5
Cyfanswm
4.15/5
MeloTTS
Free
Naturioldeb
3.8/5
Cywirdeb
4.1/5
Cyflymder
4.6/5
Cyfanswm
4.13/5
Dia TTS
Standard
Naturioldeb
4.6/5
Cywirdeb
4.3/5
Cyflymder
3.2/5
Cyfanswm
4.09/5
VITS
Free
Naturioldeb
3.4/5
Cywirdeb
4.0/5
Cyflymder
4.8/5
Cyfanswm
4.0/5
Orpheus
Standard
Naturioldeb
4.3/5
Cywirdeb
4.1/5
Cyflymder
3.5/5
Cyfanswm
4.0/5
OpenVoice
Premium
Naturioldeb
4.0/5
Cywirdeb
4.1/5
Cyflymder
3.9/5
Cyfanswm
4.0/5
IndexTTS-2
Standard
Naturioldeb
4.3/5
Cywirdeb
4.1/5
Cyflymder
3.2/5
Cyfanswm
3.91/5
Spark TTS
Standard
Naturioldeb
4.2/5
Cywirdeb
4.0/5
Cyflymder
3.4/5
Cyfanswm
3.9/5
Parler TTS
Standard
Naturioldeb
4.1/5
Cywirdeb
3.9/5
Cyflymder
3.4/5
Cyfanswm
3.83/5
Tortoise TTS
Premium
Naturioldeb
4.6/5
Cywirdeb
4.4/5
Cyflymder
1.8/5
Cyfanswm
3.7/5
Bark
Standard
Naturioldeb
4.2/5
Cywirdeb
3.8/5
Cyflymder
2.5/5
Cyfanswm
3.57/5
Methodoleg Benchmark
Gosodiad Arbrofion
- Caledwedd: 4x NVIDIA Tesla P40 (24GB VRAM pob un), 96GB cyfanswm
- Testun prawf: 5 darn safonol sy'n cwmpasu gwahanol fathau o siarad (narratif, cyfathrebu, technegol, emosiynol, aml-ieithog)
- Arfarniad: Metrigau awtomatig (amcangyfrif MOS, WER, RTF) ynghyd â phrofion gwrando dynol
- Rhedeg: Prawfwyd pob model 10 gwaith bob taith, sgôr cyfartalog
Meini prawf sgorio
- Naturioldeb (40%): Prosody, intonation, rhythm, emotions - pa mor ddynol mae'n swnio?
- Cywirdeb (30%) Cywirdeb ynganiad, cyfradd gwallau geiriau, dealladwyedd
- Cyflymder (30%): Ffactor amser real (eiliadau sain / eiliadau creu). Uwch = cyflymach.
- Cyfanswm: Cyfartaledd pwysedig: 0.4 x Naturioldeb + 0.3 x Tryloywder + 0.3 x Cyflymder
Nodyn: Mae meini prawf yn adlewyrchu perfformiad ar ein caledwedd a thestunau prawf penodol. Gall ansawdd y byd go iawn amrywio yn seiliedig ar y testun mewnbwn, yr iaith, a'r dewis llais. Mae graddau'r gymuned yn darparu arwydd cyfatebol yn seiliedig ar ddefnydd go iawn amrywiol.
Cwestiynau a Ofynnir yn Aml
Canfod eich llais perffaith
Ceisiwch unrhyw fodel am ddim gyda Kokoro, Piper, VITS, neu MeloTTS. Nid oes angen cyfrif.